ruby_llm-agents 0.5.0 → 1.0.0.beta.1

data/lib/ruby_llm/agents/core/configuration.rb

@@ -0,0 +1,981 @@
+# frozen_string_literal: true
+
+module RubyLLM
+  module Agents
+    # Global configuration for RubyLLM::Agents
+    #
+    # Provides centralized settings for agent behavior, dashboard authentication,
+    # caching, and observability thresholds.
+    #
+    # @example Basic configuration
+    #   RubyLLM::Agents.configure do |config|
+    #     config.default_model = "gpt-4o"
+    #     config.default_temperature = 0.7
+    #     config.async_logging = true
+    #   end
+    #
+    # @example Dashboard with HTTP Basic Auth
+    #   RubyLLM::Agents.configure do |config|
+    #     config.basic_auth_username = ENV["AGENTS_USER"]
+    #     config.basic_auth_password = ENV["AGENTS_PASS"]
+    #   end
+    #
+    # @example Dashboard with custom authentication
+    #   RubyLLM::Agents.configure do |config|
+    #     config.dashboard_parent_controller = "AdminController"
+    #     config.dashboard_auth = ->(controller) { controller.current_user&.admin? }
+    #   end
+    #
+    # @see RubyLLM::Agents.configure
+    # @api public
+    class Configuration
+      # @!attribute [rw] default_model
+      #   The default LLM model identifier for all agents.
+      #   Can be overridden per-agent using the `model` DSL method.
+      #   @return [String] Model identifier (default: "gemini-2.0-flash")
+      #   @example
+      #     config.default_model = "gpt-4o"
+
+      # @!attribute [rw] default_temperature
+      #   The default temperature for LLM responses (0.0 to 2.0).
+      #   Lower values produce more deterministic outputs.
+      #   @return [Float] Temperature value (default: 0.0)
+
+      # @!attribute [rw] default_timeout
+      #   Maximum seconds to wait for an LLM response before timing out.
+      #   @return [Integer] Timeout in seconds (default: 60)
+
+      # @!attribute [rw] async_logging
+      #   Whether to log executions via background job (recommended for production).
+      #   When false, executions are logged synchronously.
+      #   @return [Boolean] Enable async logging (default: true)
+
+      # @!attribute [rw] retention_period
+      #   How long to retain execution records before cleanup.
+      #   @return [ActiveSupport::Duration] Retention period (default: 30.days)
+
+      # @!attribute [rw] anomaly_cost_threshold
+      #   Cost threshold in dollars that triggers anomaly logging.
+      #   Executions exceeding this cost are logged as warnings.
+      #   @return [Float] Cost threshold in USD (default: 5.00)
+
+      # @!attribute [rw] anomaly_duration_threshold
+      #   Duration threshold in milliseconds that triggers anomaly logging.
+      #   @return [Integer] Duration threshold in ms (default: 10_000)
+
+      # @!attribute [rw] dashboard_auth
+      #   Lambda for custom dashboard authentication.
+      #   Receives the controller instance, should return truthy to allow access.
+      #   @return [Proc] Authentication lambda (default: allows all)
+      #   @example
+      #     config.dashboard_auth = ->(c) { c.current_user&.admin? }
+
+      # @!attribute [rw] dashboard_parent_controller
+      #   Parent controller class name for the dashboard.
+      #   Use this to inherit authentication from your app's admin controller.
+      #   @return [String] Controller class name (default: "ActionController::Base")
+
+      # @!attribute [rw] basic_auth_username
+      #   Username for HTTP Basic Auth on the dashboard.
+      #   Both username and password must be set to enable Basic Auth.
+      #   @return [String, nil] Username or nil to disable (default: nil)
+
+      # @!attribute [rw] basic_auth_password
+      #   Password for HTTP Basic Auth on the dashboard.
+      #   @return [String, nil] Password or nil to disable (default: nil)
+
+      # @!attribute [rw] per_page
+      #   Number of records per page in dashboard listings.
+      #   @return [Integer] Records per page (default: 25)
+
+      # @!attribute [rw] recent_executions_limit
+      #   Number of recent executions shown on the dashboard home.
+      #   @return [Integer] Limit for recent executions (default: 10)
+
+      # @!attribute [rw] job_retry_attempts
+      #   Number of retry attempts for the async logging job on failure.
+      #   @return [Integer] Retry attempts (default: 3)
+
+      # @!attribute [w] cache_store
+      #   Custom cache store for agent response caching.
+      #   Falls back to Rails.cache if not set.
+      #   @return [ActiveSupport::Cache::Store, nil]
+
+      # @!attribute [rw] default_retries
+      #   Default retry configuration for all agents.
+      #   Can be overridden per-agent using the `retries` DSL method.
+      #   @return [Hash] Retry config with :max, :backoff, :base, :max_delay, :on keys
+      #   @example
+      #     config.default_retries = { max: 2, backoff: :exponential, base: 0.4, max_delay: 3.0, on: [] }
+
+      # @!attribute [rw] default_retryable_patterns
+      #   Default patterns in error messages that indicate retryable errors.
+      #   Organized by category for easy customization.
+      #   @return [Hash<Symbol, Array<String>>] Categorized patterns
+      #   @example
+      #     config.default_retryable_patterns = {
+      #       rate_limiting: ["rate limit", "429"],
+      #       server_errors: ["500", "502", "503"],
+      #       capacity: ["overloaded"]
+      #     }
+
+      # @!attribute [rw] default_fallback_models
+      #   Default fallback models for all agents.
+      #   Can be overridden per-agent using the `fallback_models` DSL method.
+      #   @return [Array<String>] List of model identifiers to try on failure
+
+      # @!attribute [rw] default_total_timeout
+      #   Default total timeout across all retry attempts.
+      #   Can be overridden per-agent using the `total_timeout` DSL method.
+      #   @return [Integer, nil] Total timeout in seconds, or nil for no limit
+
+      # @!attribute [rw] default_streaming
+      #   Whether streaming mode is enabled by default for all agents.
+      #   When enabled and a block is passed to call, chunks are yielded as they arrive.
+      #   Can be overridden per-agent using the `streaming` DSL method.
+      #   @return [Boolean] Enable streaming (default: false)
+      #   @example
+      #     config.default_streaming = true
+
+      # @!attribute [rw] default_tools
+      #   Default tools available to all agents.
+      #   Should be an array of RubyLLM::Tool classes.
+      #   Can be overridden or extended per-agent using the `tools` DSL method.
+      #   @return [Array<Class>] Tool classes (default: [])
+      #   @example
+      #     config.default_tools = [WeatherTool, SearchTool]
+
+      # @!attribute [rw] default_thinking
+      #   Default thinking/reasoning configuration for all agents.
+      #   When set, enables extended thinking for supported models (Claude, Gemini, etc.).
+      #   Can be overridden per-agent using the `thinking` DSL method.
+      #   @return [Hash, nil] Thinking config with :effort and/or :budget keys (default: nil)
+      #   @example Enable medium-effort thinking globally
+      #     config.default_thinking = { effort: :medium }
+      #   @example Enable high-effort thinking with budget
+      #     config.default_thinking = { effort: :high, budget: 10000 }
+
+      # @!attribute [rw] budgets
+      #   Budget configuration for cost governance.
+      #   @return [Hash, nil] Budget config with :global_daily, :global_monthly, :per_agent_daily, :per_agent_monthly, :enforcement keys
+      #   @example
+      #     config.budgets = {
+      #       global_daily: 25.0,
+      #       global_monthly: 300.0,
+      #       per_agent_daily: { "ContentAgent" => 5.0 },
+      #       per_agent_monthly: { "ContentAgent" => 120.0 },
+      #       enforcement: :soft
+      #     }
+
+      # @!attribute [rw] alerts
+      #   Alert configuration for notifications.
+      #   @return [Hash, nil] Alert config with :slack_webhook_url, :webhook_url, :on_events, :custom keys
+      #   @example
+      #     config.alerts = {
+      #       slack_webhook_url: ENV["SLACK_WEBHOOK"],
+      #       webhook_url: ENV["AGENTS_WEBHOOK"],
+      #       on_events: [:budget_soft_cap, :budget_hard_cap, :breaker_open],
+      #       custom: ->(event, payload) { Rails.logger.info("Alert: #{event}") }
+      #     }
+
+      # @!attribute [rw] persist_prompts
+      #   Whether to persist system and user prompts in execution records.
+      #   Set to false to reduce storage or for privacy compliance.
+      #   @return [Boolean] Enable prompt persistence (default: true)
+
+      # @!attribute [rw] persist_responses
+      #   Whether to persist LLM responses in execution records.
+      #   Set to false to reduce storage or for privacy compliance.
+      #   @return [Boolean] Enable response persistence (default: true)
+
+      # @!attribute [rw] redaction
+      #   Redaction configuration for PII and sensitive data.
+      #   @return [Hash, nil] Redaction config with :fields, :patterns, :placeholder, :max_value_length keys
+      #   @example
+      #     config.redaction = {
+      #       fields: %w[password api_key email ssn],
+      #       patterns: [/\b\d{3}-\d{2}-\d{4}\b/],
+      #       placeholder: "[REDACTED]",
+      #       max_value_length: 5000
+      #     }
+
+      # @!attribute [rw] multi_tenancy_enabled
+      #   Whether multi-tenancy features are enabled.
+      #   When false, the gem behaves exactly as before (backward compatible).
+      #   @return [Boolean] Enable multi-tenancy (default: false)
+      #   @example
+      #     config.multi_tenancy_enabled = true
+
+      # @!attribute [rw] tenant_resolver
+      #   Lambda that returns the current tenant identifier.
+      #   Called whenever tenant context is needed for budget tracking,
+      #   circuit breakers, and execution recording.
+      #   @return [Proc] Tenant resolution lambda (default: -> { nil })
+      #   @example Using Rails CurrentAttributes
+      #     config.tenant_resolver = -> { Current.tenant&.id }
+      #   @example Using request store
+      #     config.tenant_resolver = -> { RequestStore[:tenant_id] }
+
+      # @!attribute [rw] tenant_config_resolver
+      #   Lambda that returns tenant configuration without querying the database.
+      #   Called when resolving tenant budget config. If set, this takes priority
+      #   over the TenantBudget database lookup.
+      #   @return [Proc, nil] Tenant config resolver lambda (default: nil)
+      #   @example Using an external tenant service
+      #     config.tenant_config_resolver = ->(tenant_id) {
+      #       tenant = Tenant.find(tenant_id)
+      #       {
+      #         name: tenant.name,
+      #         daily_limit: tenant.subscription.daily_budget,
+      #         monthly_limit: tenant.subscription.monthly_budget,
+      #         daily_token_limit: tenant.subscription.daily_tokens,
+      #         monthly_token_limit: tenant.subscription.monthly_tokens,
+      #         enforcement: tenant.subscription.hard_limits? ? :hard : :soft
+      #       }
+      #     }
+
+      # @!attribute [rw] persist_messages_summary
+      #   Whether to persist a summary of conversation messages in execution records.
+      #   When true, stores message count and first/last messages (truncated).
+      #   Set to false to disable message summary persistence.
+      #   @return [Boolean] Enable messages summary persistence (default: true)
+
+      # @!attribute [rw] messages_summary_max_length
+      #   Maximum character length for message content in the summary.
+      #   Content exceeding this length will be truncated with "...".
+      #   @return [Integer] Max length for message content (default: 500)
+
+      # @!attribute [rw] default_embedding_model
+      #   The default embedding model identifier for all embedders.
+      #   Can be overridden per-embedder using the `model` DSL method.
+      #   @return [String] Model identifier (default: "text-embedding-3-small")
+      #   @example
+      #     config.default_embedding_model = "text-embedding-3-large"
+
+      # @!attribute [rw] default_embedding_dimensions
+      #   The default vector dimensions for embeddings.
+      #   Set to nil to use the model's default dimensions.
+      #   Some models (like OpenAI text-embedding-3) support dimension reduction.
+      #   @return [Integer, nil] Dimensions or nil for model default (default: nil)
+      #   @example
+      #     config.default_embedding_dimensions = 512
+
+      # @!attribute [rw] default_embedding_batch_size
+      #   The default batch size for embedding operations.
+      #   When embedding multiple texts, they are split into batches of this size.
+      #   @return [Integer] Batch size (default: 100)
+      #   @example
+      #     config.default_embedding_batch_size = 50
+
+      # @!attribute [rw] track_embeddings
+      #   Whether to track embedding executions in the database.
+      #   When enabled, embedding operations are logged like agent executions.
+      #   @return [Boolean] Enable embedding tracking (default: true)
+      #   @example
+      #     config.track_embeddings = false
+
+      # @!attribute [rw] default_moderation_model
+      #   The default moderation model identifier for all agents.
+      #   Can be overridden per-agent using the `moderation` DSL method.
+      #   @return [String] Model identifier (default: "omni-moderation-latest")
+      #   @example
+      #     config.default_moderation_model = "text-moderation-007"
+
+      # @!attribute [rw] default_moderation_threshold
+      #   The default threshold for moderation scores.
+      #   Content with scores at or above this threshold will be flagged.
+      #   Set to nil to use the provider's default flagging.
+      #   @return [Float, nil] Threshold (0.0-1.0) or nil for provider default (default: nil)
+      #   @example
+      #     config.default_moderation_threshold = 0.8
+
+      # @!attribute [rw] default_moderation_action
+      #   The default action when content is flagged.
+      #   Can be overridden per-agent using the `moderation` DSL method.
+      #   @return [Symbol] Action (:block, :raise, :warn, :log) (default: :block)
+      #   @example
+      #     config.default_moderation_action = :raise
+
+      # @!attribute [rw] track_moderation
+      #   Whether to track moderation executions in the database.
+      #   When enabled, moderation operations are logged as executions.
+      #   @return [Boolean] Enable moderation tracking (default: true)
+      #   @example
+      #     config.track_moderation = false
+
+      # @!attribute [rw] default_transcription_model
+      #   The default transcription model identifier for all transcribers.
+      #   Can be overridden per-transcriber using the `model` DSL method.
+      #   @return [String] Model identifier (default: "whisper-1")
+      #   @example
+      #     config.default_transcription_model = "gpt-4o-transcribe"
+
+      # @!attribute [rw] track_transcriptions
+      #   Whether to track transcription executions in the database.
+      #   When enabled, transcription operations are logged as executions.
+      #   @return [Boolean] Enable transcription tracking (default: true)
+      #   @example
+      #     config.track_transcriptions = false
+
+      # @!attribute [rw] default_tts_provider
+      #   The default TTS provider for all speakers.
+      #   Can be overridden per-speaker using the `provider` DSL method.
+      #   @return [Symbol] Provider (:openai, :elevenlabs, :google, :polly) (default: :openai)
+      #   @example
+      #     config.default_tts_provider = :elevenlabs
+
+      # @!attribute [rw] default_tts_model
+      #   The default TTS model identifier for all speakers.
+      #   Can be overridden per-speaker using the `model` DSL method.
+      #   @return [String] Model identifier (default: "tts-1")
+      #   @example
+      #     config.default_tts_model = "tts-1-hd"
+
+      # @!attribute [rw] default_tts_voice
+      #   The default voice for all speakers.
+      #   Can be overridden per-speaker using the `voice` DSL method.
+      #   @return [String] Voice name (default: "nova")
+      #   @example
+      #     config.default_tts_voice = "alloy"
+
+      # @!attribute [rw] track_speech
+      #   Whether to track speech executions in the database.
+      #   When enabled, speech operations are logged as executions.
+      #   @return [Boolean] Enable speech tracking (default: true)
+      #   @example
+      #     config.track_speech = false
+
+      # @!attribute [rw] async_max_concurrency
+      #   Maximum number of concurrent async operations when using batch processing.
+      #   Controls the semaphore limit for Async::Semaphore.
+      #   @return [Integer] Max concurrent operations (default: 10)
+      #   @example
+      #     config.async_max_concurrency = 20
+
+      # @!attribute [rw] root_directory
+      #   The root directory name under app/ for all LLM components.
+      #   This allows customization of the directory structure.
+      #   @return [String] Directory name (default: "llm")
+      #   @example
+      #     config.root_directory = "ai"  # Creates app/ai/ instead of app/llm/
+
+      # @!attribute [rw] root_namespace
+      #   The root namespace for all LLM component classes.
+      #   This should match the root_directory in camelized form.
+      #   Set to nil or "" to use no namespace (classes at top-level).
+      #   @return [String, nil] Namespace (default: "Llm")
+      #   @example Custom namespace
+      #     config.root_namespace = "AI"  # Uses AI:: instead of Llm::
+      #   @example No namespace (top-level classes)
+      #     config.root_namespace = nil   # Uses top-level classes (ApplicationAgent, etc.)
+
+      # Attributes without validation (simple accessors)
+      attr_accessor :default_model,
+                    :async_logging,
+                    :async_max_concurrency,
+                    :retention_period,
+                    :dashboard_parent_controller,
+                    :basic_auth_username,
+                    :basic_auth_password,
+                    :default_fallback_models,
+                    :default_total_timeout,
+                    :default_streaming,
+                    :default_tools,
+                    :default_thinking,
+                    :alerts,
+                    :persist_prompts,
+                    :persist_responses,
+                    :redaction,
+                    :multi_tenancy_enabled,
+                    :persist_messages_summary,
+                    :default_retryable_patterns,
+                    :default_embedding_model,
+                    :default_embedding_dimensions,
+                    :default_embedding_batch_size,
+                    :track_embeddings,
+                    :default_moderation_model,
+                    :default_moderation_threshold,
+                    :default_moderation_action,
+                    :track_moderation,
+                    :default_transcription_model,
+                    :track_transcriptions,
+                    :default_tts_provider,
+                    :default_tts_model,
+                    :default_tts_voice,
+                    :track_speech,
+                    :track_executions,
+                    :track_audio,
+                    :track_cache_hits,
+                    :default_image_model,
+                    :default_image_size,
+                    :default_image_quality,
+                    :default_image_style,
+                    :max_image_prompt_length,
+                    :track_image_generation,
+                    :image_model_aliases,
+                    :litellm_pricing_url,
+                    :litellm_pricing_cache_ttl,
+                    :default_image_cost,
+                    :image_model_pricing,
+                    :default_variator_model,
+                    :default_editor_model,
+                    :default_transformer_model,
+                    :default_upscaler_model,
+                    :default_variation_strength,
+                    :default_transform_strength,
+                    :default_analyzer_model,
+                    :default_analysis_type,
+                    :default_analyzer_max_tags,
+                    :default_background_remover_model,
+                    :default_background_output_format,
+                    :root_directory,
+                    :root_namespace
+
+      # Attributes with validation (readers only, custom setters below)
+      attr_reader :default_temperature,
+                  :default_timeout,
+                  :anomaly_cost_threshold,
+                  :anomaly_duration_threshold,
+                  :per_page,
+                  :recent_executions_limit,
+                  :job_retry_attempts,
+                  :messages_summary_max_length,
+                  :dashboard_auth,
+                  :tenant_resolver,
+                  :tenant_config_resolver,
+                  :default_retries,
+                  :budgets
+
+      attr_writer :cache_store
+
+      # Sets default_temperature with validation
+      #
+      # @param value [Float] Temperature (0.0 to 2.0)
+      # @raise [ArgumentError] If value is outside valid range
+      def default_temperature=(value)
+        validate_range!(:default_temperature, value, 0.0, 2.0)
+        @default_temperature = value
+      end
+
+      # Sets default_timeout with validation
+      #
+      # @param value [Integer] Timeout in seconds (must be > 0)
+      # @raise [ArgumentError] If value is not positive
+      def default_timeout=(value)
+        validate_positive!(:default_timeout, value)
+        @default_timeout = value
+      end
+
+      # Sets anomaly_cost_threshold with validation
+      #
+      # @param value [Float] Cost threshold (must be >= 0)
+      # @raise [ArgumentError] If value is negative
+      def anomaly_cost_threshold=(value)
+        validate_non_negative!(:anomaly_cost_threshold, value)
+        @anomaly_cost_threshold = value
+      end
+
+      # Sets anomaly_duration_threshold with validation
+      #
+      # @param value [Integer] Duration threshold in ms (must be >= 0)
+      # @raise [ArgumentError] If value is negative
+      def anomaly_duration_threshold=(value)
+        validate_non_negative!(:anomaly_duration_threshold, value)
+        @anomaly_duration_threshold = value
+      end
+
+      # Sets per_page with validation
+      #
+      # @param value [Integer] Records per page (must be > 0)
+      # @raise [ArgumentError] If value is not positive
+      def per_page=(value)
+        validate_positive!(:per_page, value)
+        @per_page = value
+      end
+
+      # Sets recent_executions_limit with validation
+      #
+      # @param value [Integer] Limit (must be > 0)
+      # @raise [ArgumentError] If value is not positive
+      def recent_executions_limit=(value)
+        validate_positive!(:recent_executions_limit, value)
+        @recent_executions_limit = value
+      end
+
+      # Sets job_retry_attempts with validation
+      #
+      # @param value [Integer] Retry attempts (must be >= 0)
+      # @raise [ArgumentError] If value is negative
+      def job_retry_attempts=(value)
+        validate_non_negative!(:job_retry_attempts, value)
+        @job_retry_attempts = value
+      end
+
+      # Sets messages_summary_max_length with validation
+      #
+      # @param value [Integer] Max length (must be > 0)
+      # @raise [ArgumentError] If value is not positive
+      def messages_summary_max_length=(value)
+        validate_positive!(:messages_summary_max_length, value)
+        @messages_summary_max_length = value
+      end
+
+      # Sets dashboard_auth with validation
+      #
+      # @param value [Proc, nil] Authentication lambda or nil
+      # @raise [ArgumentError] If value is not callable or nil
+      def dashboard_auth=(value)
+        validate_callable!(:dashboard_auth, value, allow_nil: true)
+        @dashboard_auth = value
+      end
+
+      # Sets tenant_resolver with validation
+      #
+      # @param value [Proc] Tenant resolution lambda (must be callable)
+      # @raise [ArgumentError] If value is not callable
+      def tenant_resolver=(value)
+        validate_callable!(:tenant_resolver, value, allow_nil: false)
+        @tenant_resolver = value
+      end
+
+      # Sets tenant_config_resolver with validation
+      #
+      # @param value [Proc, nil] Tenant config resolver lambda or nil
+      # @raise [ArgumentError] If value is not callable or nil
+      def tenant_config_resolver=(value)
+        validate_callable!(:tenant_config_resolver, value, allow_nil: true)
+        @tenant_config_resolver = value
+      end
+
+      # Sets default_retries with validation
+      #
+      # @param value [Hash] Retry configuration
+      # @raise [ArgumentError] If any values are invalid
+      def default_retries=(value)
+        validate_retries!(value)
+        @default_retries = value
+      end
+
+      # Sets budgets with validation
+      #
+      # @param value [Hash, nil] Budget configuration
+      # @raise [ArgumentError] If enforcement is invalid
+      def budgets=(value)
+        validate_budgets!(value)
+        @budgets = value
+      end
+
+      # Sets default_embedding_batch_size with validation
+      #
+      # @param value [Integer] Batch size (must be > 0)
+      # @raise [ArgumentError] If value is not positive
+      def default_embedding_batch_size=(value)
+        validate_positive!(:default_embedding_batch_size, value)
+        @default_embedding_batch_size = value
+      end
+
+      # Sets default_embedding_dimensions with validation
+      #
+      # @param value [Integer, nil] Dimensions (must be nil or > 0)
+      # @raise [ArgumentError] If value is not nil or positive
+      def default_embedding_dimensions=(value)
+        unless value.nil? || (value.is_a?(Numeric) && value > 0)
+          raise ArgumentError, "default_embedding_dimensions must be nil or greater than 0"
+        end
+
+        @default_embedding_dimensions = value
+      end
+
+      # Initializes configuration with default values
+      #
+      # @return [Configuration] A new configuration instance with defaults
+      # @api private
+      def initialize
+        @default_model = "gemini-2.0-flash"
+        @default_temperature = 0.0
+        @default_timeout = 60
+        @cache_store = nil
+        @async_logging = true
+        @async_max_concurrency = 10
+        @retention_period = 30.days
+        @anomaly_cost_threshold = 5.00
+        @anomaly_duration_threshold = 10_000
+        @dashboard_auth = ->(_controller) { true }
+        @dashboard_parent_controller = "ActionController::Base"
+        @basic_auth_username = nil
+        @basic_auth_password = nil
+        @per_page = 25
+        @recent_executions_limit = 10
+        @job_retry_attempts = 3
+
+        # Reliability defaults (all disabled by default for backward compatibility)
+        @default_retries = { max: 0, backoff: :exponential, base: 0.4, max_delay: 3.0, on: [] }
+        @default_fallback_models = []
+        @default_total_timeout = nil
+        @default_retryable_patterns = {
+          rate_limiting: ["rate limit", "rate_limit", "too many requests", "429"],
+          server_errors: ["500", "502", "503", "504", "service unavailable",
+                         "internal server error", "bad gateway", "gateway timeout"],
+          capacity: ["overloaded", "capacity"]
+        }
+
+        # Streaming, tools, and thinking defaults
+        @default_streaming = false
+        @default_tools = []
+        @default_thinking = nil
+
+        # Governance defaults
+        @budgets = nil
+        @alerts = nil
+        @persist_prompts = true
+        @persist_responses = true
+        @redaction = nil
+
+        # Multi-tenancy defaults (disabled for backward compatibility)
+        @multi_tenancy_enabled = false
+        @tenant_resolver = -> { nil }
+        @tenant_config_resolver = nil
+
+        # Messages summary defaults
+        @persist_messages_summary = true
+        @messages_summary_max_length = 500
+
+        # Embedding defaults
+        @default_embedding_model = "text-embedding-3-small"
+        @default_embedding_dimensions = nil
+        @default_embedding_batch_size = 100
+        @track_embeddings = true
+
+        # Moderation defaults
+        @default_moderation_model = "omni-moderation-latest"
+        @default_moderation_threshold = nil
+        @default_moderation_action = :block
+        @track_moderation = true
+
+        # Transcription defaults
+        @default_transcription_model = "whisper-1"
+        @track_transcriptions = true
+
+        # TTS/Speech defaults
+        @default_tts_provider = :openai
+        @default_tts_model = "tts-1"
+        @default_tts_voice = "nova"
+        @track_speech = true
+
+        # Execution/conversation agent tracking
+        @track_executions = true
+        @track_audio = true
+        @track_cache_hits = true
+
+        # Image Generation defaults
+        @default_image_model = "gpt-image-1"
+        @default_image_size = "1024x1024"
+        @default_image_quality = "standard"
+        @default_image_style = "vivid"
+        @max_image_prompt_length = 4000
+        @track_image_generation = true
+        @image_model_aliases = {
+          fast: "flux-schnell",
+          quality: "gpt-image-1",
+          cheap: "sdxl"
+        }
+
+        # Image Pricing defaults
+        @litellm_pricing_url = nil  # Use default from Pricing module
+        @litellm_pricing_cache_ttl = nil  # Use default (24 hours)
+        @default_image_cost = 0.04  # Fallback cost per image
+        @image_model_pricing = {}  # User-defined pricing overrides
+
+        # Phase 2: Image Variation, Editing, Transformation, Upscaling defaults
+        @default_variator_model = nil  # Falls back to default_image_model
+        @default_editor_model = nil  # Falls back to default_image_model
+        @default_transformer_model = "sdxl"  # SDXL is good for transformations
+        @default_upscaler_model = "real-esrgan"  # Real-ESRGAN for upscaling
+        @default_variation_strength = 0.5  # How different variations should be
+        @default_transform_strength = 0.75  # How much to transform
+
+        # Phase 3: Image Analysis and Background Removal defaults
+        @default_analyzer_model = "gpt-4o"  # Vision model for analysis
+        @default_analysis_type = :detailed  # Default analysis type
+        @default_analyzer_max_tags = 10  # Max tags to extract
+        @default_background_remover_model = "rembg"  # Default remover model
+        @default_background_output_format = :png  # Output format for transparency
+
+        # Directory structure defaults
+        @root_directory = "llm"  # Root directory under app/
+        @root_namespace = "Llm"  # Root namespace for classes
+      end
+
+      # Returns the configured cache store, falling back to Rails.cache
+      #
+      # @return [ActiveSupport::Cache::Store] The cache store instance
+      # @example Using a custom cache store
+      #   config.cache_store = ActiveSupport::Cache::MemoryStore.new
+      def cache_store
+        @cache_store || Rails.cache
+      end
+
+      # Returns whether budgets are configured and enforcement is enabled
+      #
+      # @return [Boolean] true if budgets are configured with enforcement
+      def budgets_enabled?
+        budgets.is_a?(Hash) && budgets[:enforcement] && budgets[:enforcement] != :none
+      end
+
+      # Returns the budget enforcement mode
+      #
+      # @return [Symbol] :none, :soft, or :hard
+      def budget_enforcement
+        budgets&.dig(:enforcement) || :none
+      end
+
+      # Returns all retryable patterns as a flat array
+      #
+      # @return [Array<String>] All patterns from all categories
+      def all_retryable_patterns
+        default_retryable_patterns.values.flatten.uniq
+      end
+
+      # Returns whether alerts are configured
+      #
+      # @return [Boolean] true if any alert destination is configured
+      def alerts_enabled?
+        return false unless alerts.is_a?(Hash)
+
+        alerts[:slack_webhook_url].present? ||
+          alerts[:webhook_url].present? ||
+          alerts[:custom].present?
+      end
+
+      # Returns the list of events to alert on
+      #
+      # @return [Array<Symbol>] Event names to trigger alerts
+      def alert_events
+        alerts&.dig(:on_events) || []
+      end
+
+      # Returns merged redaction fields (default sensitive keys + configured)
+      #
+      # @return [Array<String>] Field names to redact
+      def redaction_fields
+        default_fields = %w[password token api_key secret credential auth key access_token]
+        configured_fields = redaction&.dig(:fields) || []
+        (default_fields + configured_fields).map(&:downcase).uniq
+      end
+
+      # Returns redaction patterns
+      #
+      # @return [Array<Regexp>] Patterns to match and redact
+      def redaction_patterns
+        redaction&.dig(:patterns) || []
+      end
+
+      # Returns the redaction placeholder string
+      #
+      # @return [String] Placeholder to replace redacted values
+      def redaction_placeholder
+        redaction&.dig(:placeholder) || "[REDACTED]"
+      end
+
+      # Returns the maximum value length before truncation
+      #
+      # @return [Integer, nil] Max length, or nil for no limit
+      def redaction_max_value_length
+        redaction&.dig(:max_value_length)
+      end
+
+      # Returns whether multi-tenancy is enabled
+      #
+      # @return [Boolean] true if multi-tenancy is enabled
+      def multi_tenancy_enabled?
+        @multi_tenancy_enabled == true
+      end
+
+      # Returns the current tenant ID from the resolver
+      #
+      # @return [String, nil] Current tenant identifier or nil
+      def current_tenant_id
+        return nil unless multi_tenancy_enabled?
+
+        tenant_resolver&.call
+      end
+
+      # Returns whether the async gem is available
+      #
+      # @return [Boolean] true if async gem is loaded
+      def async_available?
+        defined?(::Async) && defined?(::Async::Semaphore)
+      end
+
+      # Returns whether we're currently inside an async context
+      #
+      # @return [Boolean] true if running in a fiber with async scheduler
+      def async_context?
+        return false unless async_available?
+
+        defined?(::Async::Task) && ::Async::Task.current?
+      rescue StandardError
+        false
+      end
+
+      # Returns the full namespace for a given category
+      #
+      # @param category [Symbol, nil] Category (:audio, :image, :text, or nil for root)
+      # @return [String, nil] Full namespace string, or nil if no namespace configured
+      # @example With default namespace
+      #   namespace_for(:image) #=> "Llm::Image"
+      #   namespace_for(nil)    #=> "Llm"
+      # @example With no namespace (root_namespace = nil)
+      #   namespace_for(:image) #=> "Image"
+      #   namespace_for(nil)    #=> nil
+      def namespace_for(category = nil)
+        # Handle no namespace configuration
+        if root_namespace.blank?
+          case category
+          when :audio then "Audio"
+          when :image then "Image"
+          when :text then "Text"
+          else nil
+          end
+        else
+          case category
+          when :audio then "#{root_namespace}::Audio"
+          when :image then "#{root_namespace}::Image"
+          when :text then "#{root_namespace}::Text"
+          else root_namespace
+          end
+        end
+      end
+
+      # Returns the full path under app/ for a given category and type
+      #
+      # @param category [Symbol, nil] Category (:audio, :image, :text, or nil)
+      # @param type [String] Component type (e.g., "agents", "speakers", "generators")
+      # @return [String] Full path relative to Rails.root
+      # @example
+      #   path_for(:image, "generators") #=> "app/llm/image/generators"
+      #   path_for(nil, "agents")        #=> "app/llm/agents"
+      def path_for(category, type)
+        case category
+        when :audio then "app/#{root_directory}/audio/#{type}"
+        when :image then "app/#{root_directory}/image/#{type}"
+        when :text then "app/#{root_directory}/text/#{type}"
+        else "app/#{root_directory}/#{type}"
+        end
+      end
+
+      # Returns all autoload paths for LLM components
+      #
+      # @return [Array<String>] List of paths relative to Rails.root
+      def all_autoload_paths
+        [
+          # Top-level
+          path_for(nil, "agents"),
+          path_for(nil, "workflows"),
+          path_for(nil, "tools"),
+          # Audio
+          path_for(:audio, "speakers"),
+          path_for(:audio, "transcribers"),
+          # Image
+          path_for(:image, "analyzers"),
+          path_for(:image, "background_removers"),
+          path_for(:image, "editors"),
+          path_for(:image, "generators"),
+          path_for(:image, "transformers"),
+          path_for(:image, "upscalers"),
+          path_for(:image, "variators"),
+          # Text
+          path_for(:text, "embedders"),
+          path_for(:text, "moderators")
+        ]
+      end
+
+      private
+
+      # Validates that a value is within a range
+      #
+      # @param attr [Symbol] Attribute name for error message
+      # @param value [Numeric] Value to validate
+      # @param min [Numeric] Minimum value (inclusive)
+      # @param max [Numeric] Maximum value (inclusive)
+      # @raise [ArgumentError] If value is outside range
+      def validate_range!(attr, value, min, max)
+        return if value.is_a?(Numeric) && value >= min && value <= max
+
+        raise ArgumentError, "#{attr} must be between #{min} and #{max}"
+      end
+
+      # Validates that a value is positive (greater than 0)
+      #
+      # @param attr [Symbol] Attribute name for error message
+      # @param value [Numeric] Value to validate
+      # @raise [ArgumentError] If value is not positive
+      def validate_positive!(attr, value)
+        return if value.is_a?(Numeric) && value > 0
+
+        raise ArgumentError, "#{attr} must be greater than 0"
+      end
+
+      # Validates that a value is non-negative (>= 0)
+      #
+      # @param attr [Symbol] Attribute name for error message
+      # @param value [Numeric] Value to validate
+      # @raise [ArgumentError] If value is negative
+      def validate_non_negative!(attr, value)
+        return if value.is_a?(Numeric) && value >= 0
+
+        raise ArgumentError, "#{attr} must be >= 0"
+      end
+
+      # Validates that a value is callable (responds to :call)
+      #
+      # @param attr [Symbol] Attribute name for error message
+      # @param value [Object] Value to validate
+      # @param allow_nil [Boolean] Whether nil is allowed
+      # @raise [ArgumentError] If value is not callable (or nil when allowed)
+      def validate_callable!(attr, value, allow_nil:)
+        return if allow_nil && value.nil?
+        return if value.respond_to?(:call)
+
+        if allow_nil
+          raise ArgumentError, "#{attr} must be callable or nil"
+        else
+          raise ArgumentError, "#{attr} must be callable"
+        end
+      end
+
+      # Validates retries configuration hash
+      #
+      # @param value [Hash] Retries configuration
+      # @raise [ArgumentError] If any values are invalid
+      def validate_retries!(value)
+        return unless value.is_a?(Hash)
+
+        if value.key?(:backoff) && ![:exponential, :constant].include?(value[:backoff])
+          raise ArgumentError, "default_retries[:backoff] must be :exponential or :constant"
+        end
+
+        if value.key?(:base) && (!value[:base].is_a?(Numeric) || value[:base] <= 0)
+          raise ArgumentError, "default_retries[:base] must be greater than 0"
+        end
+
+        if value.key?(:max_delay) && (!value[:max_delay].is_a?(Numeric) || value[:max_delay] <= 0)
+          raise ArgumentError, "default_retries[:max_delay] must be greater than 0"
+        end
+      end
+
+      # Validates budgets configuration hash
+      #
+      # @param value [Hash, nil] Budgets configuration
+      # @raise [ArgumentError] If enforcement is invalid
+      def validate_budgets!(value)
+        return if value.nil?
+        return unless value.is_a?(Hash)
+
+        if value.key?(:enforcement) && ![:none, :soft, :hard].include?(value[:enforcement])
+          raise ArgumentError, "budgets[:enforcement] must be :none, :soft, or :hard"
+        end
+      end
+    end
+  end
+end